>All new FS's should expect to work on a "device" that is totally
>committed to the FS and looks like it starts at offset 0, and return
>EINVAL or EFAULT for out of range operations on the device.  If
>this were there now, then the MSDOSFS would be incapable of
>corrupting non-MSDOSFS disk areas with the current code.

It's always been there.  Broken code can only write outside the range
by corrupting the device number.

Bruce